Research On Classification Algorithm Of ECG Signal Recognition Based On Deep Learning

As society continues to advance and life pressures continue to rise,heart disease has become an important factor threatening human health.Electrocardiogram(ECG)characterizes the electrical activity of the human heart and can be visualized by electrocardiogram.Before the onset of heart disease,the corresponding arrhythmia is usually present in the ECG signal.Therefore,the ECG signal is identified and classified.It is of great significance for the diagnosis and treatment of heart disease.As a result,many experts and scholars have conducted extensive research on the classification and classification of ECG signals.The identification classification of ECG signals usually includes steps such as ECG signal acquisition,ECG signal preprocessing,extraction of ECG characteristics,and design identification classifier.ECG signals often contain interference noise.The presence of noise will affect the accurate detection of R wave peaks of ECG signal characteristics at the later stage,and affect the classification rate of ECG signals.Therefore,the pretreatment of ECG signals is the key to ECG signal classification.step.For the existing filtering algorithm,the noise remaining in the ECG signal after filtering is not considered.According to the difference between the ECG signal and the noise in frequency,the information in the MIT-BIH arrhythmia database is taken as the object,firstly,the MIT-BIH heart The 50 Hz power frequency interference in the electrical database is designed to filter by the finite impulse response(FIR)notch filter.The median filter is used to filter the baseline drift,and the Butterworth low-pass filter is filtered for the myoelectric interference design.In order to ensure that the noise remaining in the ECG signal is effectively filtered out,after the digital filtering,the ECG signal is double-filtered by the wavelet unbiased risk estimation threshold method,and the 3-dB coefficient is decomposed on the ECG signal by the“db5” wavelet.It is found that the residual noise belongs to high frequency noise,which is mainly distributed on the wavelet detail coefficient.The unbiased risk estimation threshold is used to threshold the decomposed wavelet detail coefficient and reconstruct the ECG signal.The research results show that the double filtering method is adopted in this paper.After filtering the ECG signal,the signal-to-noise ratio(SNR)and mean square error(MSE)are significantly improved,and a good filtering effect is achieved without causing distortion of the ECG signal.Aiming at theexisting ECG signal feature extraction algorithm,only some features in the ECG signal can be extracted and the ECG feature information can not be fully expressed,so that the classifier can not obtain a good recognition and classification effect on the ECG signal.In the ECG signal,all the feature information is completely retained in the complete heart beat.In this paper,the adaptive differential threshold method is used to accurately detect the R peak.In order to prevent the occurrence of random spikes,the dynamic adaptive threshold is improved in time to avoid the missed detection and error of the R peak.Check.A total of 106,428 groups of heartbeat signals(normal ECG signals,left bundle branch block,right bundle branch block,ventricular premature beat,pacing heartbeat)were collected from the database.In recent years,artificial intelligence technology has developed rapidly and is increasingly used in the field of biomedical engineering.In this paper,the deep learning algorithm in artificial intelligence is used to identify and classify ECG signals to improve the recognition rate and real-time monitoring of ECG signals.In this paper,four deep neural networks,such as Deep Confidence Neural Network(DBN),Convolutional Neural Network(CNN),Long and Short Memory Neural Network(LSTM),and Convolutional Recurrent Neural Network(CRNN),are designed in the MIT-BIH ECG database.Five common ECG signals(normal ECG signal,left bundle branch block,right bundle branch block,ventricular premature beat,pacing heartbeat)were identified and classified.For the first time,the fusion network model CRNN was applied to ECG signals.Identification classification.In this paper,a one-fold "cross-validation method" is used to train and test the performance of neural network to identify ECG signals.The 106,428 heart beats obtained by the MIT-BIH database are randomly divided into 70,000 training samples and 36,428 test samples.The training group data is input into the corresponding neural network for training,and when the network error reaches the preset error value,the network training ends.The efficiency of ECG signal recognition by each neural network was tested.The results showed that the overall recognition rate of ECG signals by the four deep neural networks reached over 95%,indicating that the four deep neural networks have good recognition and classification of ECG signals.Performance,especially CRNN has the best recognition effect on ECG signals,the overall recognition rate reaches 98.81%,and its generalization ability and convergence are better.Through the analysis and discussion of the influence of the number of layers of CRNN on the recognition rate,the results show that the number of network layers is not as deep aspossible.It is necessary to select a reasonable number of layers to make the network easy to train and converge,and the classification effect is best.Deep learning algorithms such as CRNN can be well applied to the automatic recognition classification of ECG signals.